APN Oriented QoE Evaluation And Optimization Scheme Design

The commercialization of 4G and 5G networks has led to rapid development in both types and quantities of network applications.The number and types of services carried by the network are becoming increasingly complex,which has put forward higher requirements for network service providers.The current industry’s demand for the internet is to have the ability to provide differentiated services and ensure the Quality of Experience(QoE).However,traditional network architecture decouples user applications from the network,making it difficult for the network to effectively utilize user side demand information and provide differentiated QoE protection for users.In order to solve the problem of network and user decoupling,industry has proposed Application-aware Networking(APN)technology.APN extends the ability to express requirements to the user side,allowing the user application to re couple with the network.As a result,users have the ability to express their own network service needs.The network side can use the QoE prediction model to provide differentiated protection for users.Although significant progress has been made in the research of QoE prediction models in the industry,existing QoE prediction models still need to be improved in terms of time dimension considerations.That is,the prediction analysis of QoE only focuses on the current system parameters,ignoring the sustained impact of system parameters in the time dimension.This QoE prediction model lacking time dimension considerations cannot effectively cope with the impact of network dynamic fluctuations on QoE when driving APN,and cannot leverage the advantages of APN in providing fine-grained differentiated services.In addition,APN is unable to effectively utilize APN parameters to describe user needs to the network,which is also an open issue left by APN proponents to the industry.Through collaborative research with experts from the IETF APN working group from Huawei,this article addresses the existing problems of QoE prediction models in APN scenarios and completes the following work:1)This article proposes a QoE time aftereffect model based on time slicing algorithm to address the limitations of existing QoE prediction models,solving the problem of lacking time dimension considerations in existing models.The model proposed in this article divides the prediction of QoE into two parts:an application performance level model and a time slicing model.The application performance level model provides the mapping relationship between the objective parameters of the network and the performance of the network application.The time slicing model is responsible for introducing the persistence effect of the time dimension and completing the time aftereffect prediction of QoE.2)This article recruits experimenters for live person testing and crowdsourcing experiments,and provides measured data for specific network application scenarios.Further fitting of the data obtained the parameters of the performance level model and the time slicing model,and classification analysis of user feature information obtained the user’s differentiated acceptable boundary parameters.Through partial autocorrelation analysis,it was proven that the user QoE sequence has time aftereffect.3)This article constructs an APN-QoE prototype system based on the QoE time aftereffect model in a simulation environment.By writing the packet processing strategy for P4 virtual switch,the forwarding base required for APN was achieved.Simulation experiments have verified the advantages of the QoE time aftereffect model compared to traditional QoE models in dynamic change networks.The results indicate that the QoE time aftereffect model can predict QoE levels by multidimensional evaluation of network status,user characteristics,and other parameters.Compared to traditional models,APN can better drive the QoE level of users in response to network changes.